Tech Talk #15 – Approximate CNN Topologies: A survey

Την Παρασκευή 12/3/2021 και ώρα 10:00 EET (Greek Time) θα έχουμε την παρουσίαση για τα Fridays Tech Talks από τoν υποψήφιo διδάκτορα Γιώργο Αρμενιάκο με τίτλο:

“Approximate CNN Topologies: A survey”

Η παρουσίαση θα γίνει μέσω webex στο παρακάτω link Webex.

Ακολουθεί μία σύντομη περίληψη της παρουσίασης:

“Convolutional Neural Networks (CNNs) are becoming popular because of their high performance and accuracy in various cognitive tasks in Machine Learning (ML). However, applications using deep neural networks are highly computationally intensive and, therefore, the need to reduce their energy consumption and demands in memory is rising. Approximate Computing (AC) has emerged as a means for improving the performance and efficiency of computing systems. The fact that deep CNNs are inherently error resilient makes Approximate Computing a solution to their high computation and storage demands. This presentation aims to provide a comprehensive survey and a comparative evaluation of recently developed approximate techniques used in (deep) CNNs. We will first introduce some key properties of convolutional neural networks which are exploited by different approximate methods. Then, we will analyze these approximation techniques and discuss their advantages and limitations.”